Differential gene expression between STING-based subtypes (C1 vs. C2) was analyzed using DESeq2 (v1.38.3). Functional enrichment was performed via clusterProfiler (v4.0) using annotations from the Gene Ontology Consortium (http://geneontology.org) and KEGG Pathway Database (https://www.genome.jp/kegg/), with significance defined as a Benjamini-Hochberg adjusted P-value <0.05. TME scores (immune/stromal) and tumor purity were calculated using the ESTIMATE algorithm (v1.0.13; http://bioinformatics.mdanderson.org/estimate/). Immune cell infiltration was deconvoluted with CIBERSORTx (https://cibersortx.stanford.edu) using the LM22 signature matrix and 1,000 permutations. Mutation profiles were processed via Maftools (v2.16.0) with TCGA somatic mutation data. All statistical thresholds are explicitly stated in figure legends. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed to interpret the functional significance of the identified genes. GO analysis (http://geneontology.org) was utilized to categorize genes into biological processes, molecular functions and cellular components. KEGG pathway analysis (https://www.genome.jp/kegg/) was used to identify significantly enriched signaling and metabolic pathways. The analyses were performed using the clusterProfiler R package (v4.0), with a significance threshold of an adjusted P-value <0.05.

Curated Cancer Cell Atlas (www.weizmann.ac.il/sites/3CA/) was used to identify the expression of STING in the TME. STRING (cn.string-db.org) and Genemiania (genemania.org) were used to identify 26 STING pathway-related genes (Table SI). The association between STING pathway-related genes and the prognosis of patients with PAAD was evaluated via univariate cox regression. A least absolute shrinkage and selection operator (LASSO) Cox regression analysis was used for constructing the signature for STING pathway-related genes. Bioinformatics analysis of PAAD transcriptomic data from public cohorts (TCGA and GEO) revealed that the majority of STING-related genes were highly expressed in PAAD (significance defined as |log2(fold change)|>1 and adjusted P-value <0.05). Based on the expression levels of these genes, PAAD patients from these cohorts were stratified into high-risk and low-risk groups. These STING-related genes exerted differential impacts on tumor mutational burden, immune infiltration and response to immunotherapy. Prognostic STING pathway-related genes were then selected via stepwise multivariate Cox regression using the Akaike information criterion for model optimization. Utilizing these genes, a prognostic model was constructed (hazard ratio, 2.639; P<0.001) and externally validated. Risk Score= ∑k=0ncoep (k)*x(k), where coef(k) and x(k) represented the regression coefficients. The groups were divided into high- and low-risk groups using the median risk score (1.476) as the cut-off. Univariate and multivariate Cox regression were used to evaluate independent factors, including clinical features and the risk score in PAAD. The Tumor Immune Dysfunction and Exclusion (TIDE) computational platform (https://tide.dfci.harvard.edu) was used to predict response to immunotherapy in high- and low-risk PAAD patient groups stratified by the STING-signature risk score.

The GEPIA (http://gepia2.cancer-pku.cn/), Cbioportal (www.cbioportal.org), HPA (www.proteinatlas.org) and TISIDB (cis.hku.hk/TISIDB/index.php) databases were used to evaluate gene expression, survival and clinical stage and grade (21) in patients with PAAD.

Lentiviral vectors expressing short hairpin RNAs (shRNAs) targeting interferon-induced protein with tetratricopeptide repeats (IFIT2), or corresponding scrambled control shRNAs (sh-NC), were constructed using the pLKO.1-puro backbone (cat. no. 8453; Addgene, Inc.). The shRNA sequences (Table SII) were designed using the Broad Institute TRC portal (v2.0) and validated by BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi) to ensure target specificity. For overexpression, full-length zinc finger DHHC-type containing 1 (ZDHHC1) complementary DNA (OriGene Technologies, Inc.) was cloned into the pCDH-CMV-MCS-EF1-Puro vector (cat. no. CD510B-1; System Biosciences, LLC). In the ZDHHC1 overexpression experiment, cells transfected with the empty vector (pCDNA3.1) served as the negative control to confirm that any observed phenotypic changes were attributable to the expression of the target gene rather than the vector itself. All constructs were verified by Sanger sequencing. Lentiviral particles were produced using a second-generation system by co-transfecting 293T cells [American Type Culture Collection (ATCC)] cultured at 37°C with the transfer vector (shRNA or overexpression plasmid), packaging plasmid psPAX2 (cat. no. 12260; Addgene, Inc.) and envelope plasmid pMD2.G (cat. no. 12259; Addgene, Inc.) at a 4:3:1 ratio (6:4.5:1.5 µg) using Lipofectamine^® 3000 (cat. no. L3000015; Thermo Fisher Scientific, Inc.). Following a 48- to 72-h transfection period, viral supernatants were harvested, filtered through 0.45-µm PVDF membranes, and concentrated by ultracentrifugation (25,000 × g, at 4°C for 2 h). Panc02 cells were seeded in 6-well plates (5×10⁴ cells/well) and incubated until 40% confluency. Cells were transduced with a multiplicity of infection of 15 used to infect cells with lentivirus in the presence of 8 µg/ml polybrene (cat. no. H9268; MilliporeSigma). After 24 h, the medium was replaced with fresh complete DMEM. Transduced cells were selected using 5 µg/ml puromycin (cat. no. ant-pr-1; InvivoGen) for 72 h. Positively selected cells were then maintained in 2 µg/ml puromycin-containing medium for all subsequent experiments to ensure stable expression. Knockdown or overexpression efficacy was validated 7 days post-selection.

Panc02 cells (ATCC CRL-2553™) were lysed in RIPA buffer (25 mM Tris-HCl pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate and 0.1% SDS) supplemented with protease inhibitor cocktail (cat. no. 11836170001; Roche Diagnostics, GmbH). Protein concentration was determined by BCA assay. Samples (40 µg/lane) were denatured in Laemmli buffer at 100°C for 10 min, separated on 10% gels using SDS-PAGE (GenScript), and transferred to PVDF membranes (cat. no. IPVH00010; MilliporeSigma). Membranes were blocked with 5% non-fat skimmed milk (cat. no. 1706404; Bio-Rad Laboratories, Inc.) in 0.1% TBST for 1 h at 25°C, followed by overnight incubation at 4°C with the following primary antibodies: Anti-IFIT2 (1:1,000; cat. no. ab305231; Abcam), anti-ZDHHC1 (1:1,000; cat. no. ab223042; Abcam) and anti-GAPDH (1:1,000; cat. no. CST 2118; Cell Signaling Technology, Inc.). After TBST washes, membranes were incubated with HRP-conjugated goat anti-rabbit IgG (1:100,000; cat. no. 31460; Invitrogen; Thermo Fisher Scientific, Inc.) for 1 h at 25°C. Signals were developed using SuperSignal™ West Pico PLUS substrate (cat.no. 34577; Thermo Fisher Scientific, Inc.) and imaged with a ChemiDoc™ MP System (Bio-Rad Laboratories, Inc.) controlled by Image Lab™ Software v6.1. Band intensity was quantified by built-in densitometry tools.

In a standard 96-well plate, cells were seeded at 5,000 Panc02 cells per well (in 100 µl DMEM). The cells were then incubated in a 37°C, 5% CO₂ cell culture incubator for 24 h. Having pre-calculated the corresponding seeding volume per well, CCK-8 solution (cat. no. CK04-13; Dojindo Laboratories, Inc.) was added to each well at 10% of the total volume of the medium, and the reaction proceeded in the dark for 1 h. The difference between the OD values of the treatment group and sh-NC or vehicle was calculated, then divided by the difference between the OD values of the control cells. The resulting value was multiplied by 100% to obtain the cell survival rate.

Cells were seeded onto sterile glass coverslips placed in 24-well culture plates at a density of 5×10⁴ cells per well and allowed to adhere for 24 h. After adherence, cells were washed once with phosphate-buffered saline (PBS) and fixed with 4% paraformaldehyde in PBS at room temperature for 30 min. Fixed cells were washed three times with PBS (5 min per wash). Cell membranes were permeabilized by incubation with pre-chilled PBS containing 0.1% Triton X-100 on ice for 2 min, followed by two washes with PBS. The TUNEL assay was performed using the In Situ Cell Death Detection Kit, Fluorescein (cat. no. 11684795910; Roche Diagnostics, GmbH). The TUNEL reaction mixture was prepared according to the manufacturer's protocol by combining 50 µl Enzyme Solution (terminal deoxynucleotidyl transferase) with 450 µl Label Solution (fluorescein-dUTP). This mixture was applied to cover the samples, which were then incubated at 37°C in a humidified dark chamber for 60 min. Unbound reagents were removed by washing three times with PBS (5 min per wash). Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) at 1 µg/ml in PBS for 10 min at room temperature in the dark, followed by two final washes with PBS. Samples were mounted using ProLong Gold Antifade Mountant (cat. no. P36930; Thermo Fisher Scientific, Inc.). Apoptotic cells (green fluorescence) and total nuclei (blue fluorescence) were visualized using an Olympus BX53 fluorescence microscope (Olympus Corporation). Quantification was performed by counting TUNEL-positive cells and total DAPI-stained nuclei in ≥10 randomly selected fields of view per sample at ×200 magnification.

For the statistical analysis component of this study, R software (version 4.3.2; The R Foundation for Statistical Computing) was used. This open-source software was downloaded from the Comprehensive R Archive Network (https://cran.r-project.org/). Continuous variables (expressed as mean ± standard deviation upon confirmation of normality by Shapiro-Wilk test) and categorical variables (reported as frequencies with proportions) were compared through multivariable regression models adjusted for demographic/clinical confounders; statistical significance was defined as bidirectional P<0.05 with Benjamini-Hochberg correction for multiple comparisons. Survival endpoints were evaluated via stratified log-rank tests and hazard ratios (95% confidence intervals). Data from CCK-8, TUNEL and western blots were treated as continuous variables. Differences between two experimental groups were assessed exclusively by unpaired two-tailed Student's t-tests. Differences between multiple experimental groups and a single control group were assessed using one-way ANOVA followed by Dunnett's multiple comparisons test. To evaluate the monotonic relationship between risk score and dendritic cells activated score, Spearman's rank-order correlation analysis was employed. This non-parametric method was selected since the data were either ordinal or continuous but failed to meet the assumption of normality, as assessed by the Shapiro-Wilk test. P<0.05 was considered to indicate a statistically significant difference.

To elucidate the clinical significance of the STING pathway in PAAD, its expression profile and prognostic relevance were systematically analyzed. Analysis of the Curated Cancer Cell Atlas revealed predominant STING expression within tumor cells, macrophages, T cells and endothelial cells of the PAAD TME (Fig. 1A) (The Cancer Genome Atlas PAAD cohort; http://portal.gdc.cancer.gov/projects/TCGA-PAAD). Utilizing GEPIA2, significantly elevated STING expression was identified in PAAD compared with normal tissue (Fig. 1B), which was associated with a poor patient prognosis (Fig. 1C) and with progressively with advancing clinical stage (22) (Fig. 1D). Genomic analysis demonstrated that STING alterations in PAAD primarily involved gene amplification (Fig. 1E). Immunohistochemical validation via The Human Protein Atlas database (https://www.proteinatlas.org) confirmed stronger STING protein staining in PAAD tissues compared with that in normal pancreatic ductal epithelium, with STING predominantly localized within cancer nests (Fig. 1F) Collectively, these findings established STING as an adverse prognostic factor in PAAD.

Given the prognostic relevance of STING, its core signaling network was characterized to define molecular subtypes. Protein-protein interaction analysis (performed using STRING database v12.0; cn.string-db.org) identified 26 STING signaling pathway-associated genes (Pearson r>0.6; all correlations significant at P<0.001 after Benjamini-Hochberg correction) (Fig. 2A and B). Unsupervised clustering based on the expression profiles of these genes stratified patients with PAAD (TCGA cohort) into two distinct molecular subtypes: Cluster 1 (C1, STING-high) and Cluster 2 (C2, STING-low) (Fig. 2C). Genes within the STING pathway were significantly upregulated in the C1 subtype (Fig. 2D). Survival analysis confirmed that patients with the C1 subtype exhibited significantly worse overall survival (Fig. 2E).

Differential gene expression analysis between C1 and C2 subtypes revealed distinct transcriptional profiles (Fig. 3A and B). Gene Ontology (GO) enrichment analysis of differentially expressed genes highlighted associations with ‘adaptive immune response’, ‘immunoglobulin complex’ and ‘production of molecular mediator of immune response’ (Fig. 3C). Kyoto Encyclopedia of Genes and genomes (KEGG) pathway analysis implicated ‘cytokine-cytokine receptor interaction’, ‘hematopoietic cell lineage’ and ‘primary immunodeficiency’ (Fig. 3D). Mutation profiling demonstrated significantly higher tumor mutational burden in the C1 subtype compared with that in the C2 subtype, characterized by frequent mutations in key oncogenes (GTPase KRas, TP53-binding protein 1, mothers against decapentaplegic homolog 4, tumor suppressor ARF and titin) (Fig. 3E and F). Evaluation of TME composition using ESTIMATE revealed significantly higher immune and stromal scores and lower tumor purity in C1 vs. C2 (Fig. 4A-D). CIBERSORTx deconvolution further indicated elevated infiltration of T cells γ δ in the C1 subtype (Fig. 4E). Consistent with enhanced immunogenicity, C1 exhibited significantly higher expression of human leukocyte antigen family genes and immune checkpoint molecules relative to C2 (Fig. 4F and G).

A risk prediction model was constructed to quantify the prognostic heterogeneity associated with STING pathway activity. Univariate Cox regression analysis identified 10 genes among the 26 STING pathway genes that were significantly associated with PAAD prognosis (Fig. 5A). LASSO Cox regression refined these into a 6-gene prognostic signature [deltex E3 ubiquitin ligase 4, interferon γ inducible protein 16, mitochondrial antiviral-signaling protein (MAVS), protein kinase DNA-activated catalytic subunit, ZDHHC1 and IFIT2] (Fig. 5B and C). This signature effectively stratified patients with PAAD into high- and low-risk groups. Survival outcomes were significantly different in the TCGA discovery cohort (Fig. 5D) and the independent GSE224564 validation cohort (Fig. 5E). Expression patterns of the 6 signature genes were consistent across both TCGA (Fig. 5F) and validation cohort (Fig. 5G) datasets. Notably, multivariate Cox regression analysis confirmed both the prognostic risk score and N stage as independent predictors of survival in patients with PAAD (Fig. 6A and B).

The STING pathway-derived prognostic risk score demonstrated significant biological relevance, showing a positive correlation with DC activation as determined by Spearman's rank correlation analysis (Fig. 6C). Assessment using the TIDE algorithm revealed significantly higher scores in high-risk vs. low-risk patients, indicating greater immune evasion potential and reduced likelihood of response to immune checkpoint blockade (Fig. 6D). Analysis of spatial expression (GSE141017) confirmed that the 6 signature genes were predominantly expressed within tumor cells of the PAAD TME (Fig. 6E and F).

Given the established prognostic relevance of STING in PAAD (characterized by high expression correlating with poor patient outcomes in the present study), its core signaling network was characterized to define molecular subtypes. While MAVS demonstrated strong prognostic associations, its functional role in PAAD has been extensively explored in prior literature (23). By contrast, ZDHHC1 and IFIT2, which also showed significant prognostic value (Fig. S1), have received considerably less attention in PAAD research. These genes were therefore prioritized for validation, observing distinct expression dynamics: IFIT2 expression significantly increased with advancing AJCC 8th edition tumor grade/stage (22) up to stage III/grade 3 before declining in advanced disease, associated with a poor prognosis, while ZDHHC1 expression decreased progressively with tumor progression and was associated with favorable outcomes (Fig. S1A-E). Immunohistochemical validation using The Human Protein Atlas (https://www.proteinatlas.org) confirmed these protein patterns in PAAD tissues (Fig. 7A). Overexpression of ZDHHC1 was confirmed in the Panc-02 cells line (Fig. 7B). Based on the confirmed superior knockdown efficiency of sh2, the stable cell line expressing sh2 was selected for all subsequent functional experiments (Fig. 7C). Functional studies in Panc02 cells revealed that overexpression of ZDHHC1 inhibited proliferation (Fig. 7D), as did knockdown of IFIT2 (Fig. 7E). Overexpression of ZDHHC1 (Fig. 7F) and knockdown of IFIT2 (Fig. 7G) both promoted apoptosis. These results established IFIT2 as a potential oncogene and ZDHHC1 as a potential tumor suppressor within the STING pathway in PAAD.